✳ the wire · analysis
Qwen 4 teaser: Alibaba posts "The becoming begins", with no model, date or benchmarks yet
An email when Qwen 4 Max answers on a public API, about 15 minutes after members, plus every frontier launch · no card

Alibaba's official Qwen account posted a teaser at 12:00 UTC on November 6: "The becoming begins." It came with no model name, release date, weights or benchmarks. At 20:56 UTC no Qwen 4 model is listed on OpenRouter, models.dev or Qwen's Hugging Face page, and Qwen Chat's top model is still Qwen3.8-Max.
WHAT THE TEASER SHOWS
A flock of birds over the sea at dusk, the Qwen logo, and the Chinese line 北冥有鱼 其名为_ ("In the northern sea there is a fish; its name is _"). It is the opening of the Zhuangzi, where the fish Kun turns into the giant bird Peng, and the name is left blank. Neither the post nor the image says "Qwen 4". A "Qwen 4" logo graphic shared widely this afternoon did not come from Alibaba's post.
WHAT ALIBABA HAS CONFIRMED
- September 22, at its Apsara Conference: Qwen 4 is in training, and Qwen 4.5 and Qwen 5 are planned at 5 to 10 trillion parameters.
- The same announcement reported recursive self-improvement (RSI) results on the current model, Qwen3.8-Max: about a month of automated runs, 33 cycles, and an Artificial Analysis score that Alibaba says rose from 40 to 45. It also described a chip-design run of more than 60 hours and more than 10,000 EDA tool calls that Alibaba says cut chip area by 42%.
- Alibaba has not said how much of Qwen 4's training uses RSI.
WHAT IS CIRCULATING (UNCONFIRMED)
- @Mr_Salio, @LuminaBench and TestingCatalog read the teaser as Qwen 4. They expect four models, Max, Plus, Flash and an open-weight 27B, with the 27B first and possibly today. Alibaba has not named any of them.
- @0x0SojalSec posted that Qwen 4 beats GPT-6 and Fable 5 and that leaked samples approach Astra and Opus 5.5 quality. The post cites no benchmark, and no public leaderboard scores a Qwen 4 model. It also says Alibaba confirmed heavy RSI use for Qwen 4 a week ago. The RSI statement we can find is Alibaba's September 22 announcement, and it is about Qwen3.8-Max.
- Earlier signs, all unconfirmed: LMArena checkpoints reported under the codenames Kalmia (Oct 22) and Kerria (Oct 27), and an attendee at Alibaba Cloud's Amsterdam event quoting CTO Feifei Li on October 29: "Qwen4 next week". Alibaba's open-source Qwen Code tool has listed qwen4, qwen4-max, qwen4-plus and qwen4-coder-plus model names in its tests since October 28.
WHAT WE DO NOT KNOW
Which model the teaser is for, when it ships, its sizes and prices, the license of any open weights, and how it scores. ArtificialWatch's radar is armed for Qwen 4 Max, Qwen 4 Flash and Qwen 4 27B, and alerts the moment any of them is listed.
Source: Qwen (Alibaba) on X ↗ · Qwen 4 Max tracker · the bench index
Cite this page
Writing about this story? Copy a link or a citation.
<a href="https://artificialwatch.com/wire/qwen-4-teaser-the-becoming-begins-what-alibaba-confirmed">Qwen 4 teaser: Alibaba posts "The becoming begins", with no model, date or benchmarks yet</a>ArtificialWatch, "Qwen 4 teaser: Alibaba posts "The becoming begins", with no model, date or benchmarks yet", https://artificialwatch.com/wire/qwen-4-teaser-the-becoming-begins-what-alibaba-confirmed

